- Rating
- 3
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
The paper looks at probe-based safety monitoring for omni-models and asks how much labeled safety data is needed across modality configurations. It defines three regimes by coverage. The authors build omni versions of WildGuardMix and Aegis 2.0 by rendering prompts into all seven configurations, and evaluate linear and MLP probes on five open omni-models. They find that the gap between O(1) and O(n) is usually larger than the gap between O(n) and O(2ⁿ), and that adding a single text+image+audio set to O(n) closes most of the latter. To address the modality gap under O(1), they learn a modality-shared, safety-informed subspace from unlabeled multimodal data plus text safety data, then train a text-only probe in that space. This gives around 3-4 AUC points on average and transfers across datasets
Critical strengths
- The O(1)/O(n)/O(2ⁿ) framing is clean and the modality-gap vs. combination-gap decomposition is a useful way to think about where supervision actually matters.
- The method is simple and practical. It needs no non-text safety labels, is learned once from general data, and the continual-updating experiment is a realistic deployment scenario.
Critical weaknesses
- The motivation of O(1) is that non-text safety data is expensive, yet the benchmark itself is generated automatically from text. In this setting O(2ⁿ) supervision is essentially free, so the obvious baseline is simply to render your text safety set into every configuration and train on that. The paper never addresses this.
- A simple per-modality mean-shift baseline is missing.
Questions for authors
- Given the datasets are generated from text automatically, why wouldn’t a practitioner just generate the full O(2ⁿ) training set? What does the O(1) method offer in that case?
- Does the learned subspace still help when the test-time rendering differs (different TTS voice, noise, real images, different fonts)?
- How do you set the decision threshold for non-text inputs in practice?
4 comments
Response 2/2
Q3. How is the decision threshold set for non-text inputs?
A3. Our main metric is AUROC (AUC), which is threshold-free, so the main results do not depend on any threshold choice. For the complementary threshold-based metrics in the Appendix (Accuracy and F1), we use a fixed threshold of 0.5 for all configurations. Even with this simple choice, the gains carry over, as shown in Table 36 in the revised manuscript (mean accuracy over the seven test configurations). Our method improves accuracy over the original monitor in 9 of 10 settings. In practice, a practitioner can calibrate the threshold once on a small multimodal validation set and reuse it. For example, calibrating to a target false positive rate requires only benign examples, which can be drawn from general (non-safety) data in each modality.
W2. Missing a per-modality mean shift baseline.
A4. Following the Reviewer's suggestion, we add a per-modality mean-shift baseline: for a test input from configuration , we use , where is the mean representation of configuration . Note that we compute from each test configuration's own training-set representations, so this baseline requires data from all configurations, which is more than our method has access to. Even so, it underperforms our -subspace method in all settings except Qwen2.5-Omni-7B, where the two are comparable. We note that for the linear probe, the mean shift leaves per-configuration AUC unchanged (identical to the original row in Table 1) as a per-configuration constant shift adds a constant to every score within that configuration and therefore does not alter the ranking. This comparison has been added to Appendix C.3.1 of the revised manuscript.
| Qwen2.5-omni-3B | Qwen2.5-omni-7B | Qwen3-omni-30B | Phi4-mm | InteractiveOmni | |
|---|---|---|---|---|---|
| Mean shift () - Linear Probe | 0.812±0.007 | 0.874±0.005 | 0.808±0.013 | 0.758±0.007 | 0.869±0.007 |
| Ours () - Linear Probe | 0.852±0.002 | 0.874±0.003 | 0.870±0.002 | 0.798±0.009 | 0.879±0.006 |
| Mean shift ()- MLP Probe | 0.830±0.007 | 0.881±0.003 | 0.819±0.010 | 0.757±0.007 | 0.876±0.004 |
| Ours () - MLP Probe | 0.869±0.001 | 0.879±0.005 | 0.869±0.002 | 0.823±0.006 | 0.887±0.005 |
The rebuttal addresses my main concerns, particularly by clarifying the practical cost advantage of the O(1) approach and adding the requested mean-shift baseline and robustness evaluations. I find the responses satisfactory and will maintain my recommendation of acceptance.
Thank you for your response! We are pleased that your concerns have been satisfactorily addressed. We appreciate your time and valuable feedback, which have helped us improve the manuscript!
Response 1/2
Thank you very much for your time and effort in providing valuable comments and for acknowledging the O(1)/O(n)/O(2n) framing, the gap decomposition, and the simple and practical method. Our responses to your concerns are summarized as follows:
Q1 / W1. Cost comparison between O(1) and O(2n)
A1. Thanks for raising the question! We would like to clarify that O(2n) supervision is not free in practice. We analyze three cost components: compute cost, API cost, and update cost.
(1) Representation-extraction compute cost scales as 2n−1. Extracting representations for all modality configurations is a major training cost. On a single 80GB A100, extraction for 10,000 prompts takes 2 GPU-hours for O(1) (text only) versus 30 for O(2n) (all seven configurations), as non-text inputs require more backbone tokens and modality encoders. In practice, a recent Anthropic report of scaling laws for safety probes [1] finds that difficult datasets require at least 1,000,000 examples to reach the error floor. Extrapolating to this scale gives 200 versus 3,000 GPU-hours per model per update, or $400 versus $6,000 at $2/A100 GPU-hour. Additional modalities, such as video, grow this cost exponentially.
(2) API cost for multimodal data generation. Combined-modality examples require an LLM call to distribute semantic content across modalities, plus image rendering and/or TTS. Closed-source APIs typically produce higher-quality images and audio than open-source alternatives, but incur generation costs that scale as O(2n) across modality configurations.
(3) Update cost for new attack types. Safety data is not static as new attack types and taxonomies appear regularly. Each new text dataset would have to be re-converted to all configurations and re-extracted, so costs (1) and (2) recur with every update.
Our O(1) subspace is estimated once from general multimodal data (which is of the kind already used in omni-model training and evaluation) and reused across monitors and updates, so it incurs only a one-time compute cost.
[1] Wu, Y., & Wei, J. (2026). Scaling Laws for Safety Probes. arXiv preprint arXiv:2610.30678.
Q2. The learned subspace's robustness to differences in test-time rendering
A2. Thanks for this constructive suggestion! Following the Reviewer's suggestion, we test robustness to different TTS voices, added noise, real images, and different fonts. All results are AUC of our O(1)-subspace method with the subspace and probe kept fixed, i.e., estimated and trained on the original rendering only.
As shown in Figure 5 in the revised manuscript, voices and noise have little effect on AUC (Δ<0.02), and fonts have almost none (Δ<0.005). Replacing typographic images with real images causes the largest drop (Δ≈0.08–0.10), reflecting an expected within-modality shift because the image component of subspace estimation data is solely from typographic images. We deliberately use typographic rendering to preserve semantic content and isolate the cross-modal gap (L205–211). For deployment, including real images in the general data used for subspace estimation may reduce this within-modality gap without additional safety labels. These results and analysis have been added to Section 4.3 of the revised manuscript.